10 December 2022 in Engineering Applications of Artificial Intelligence
21 September 2023 in The Journal of Supercomputing
15 November 2023 in IEEE Geoscience and Remote Sensing Letters
Over the past few years, there have been significant advancements in deep learning technology, leading to remarkable progress in the field of image analysis. However, when it comes to handling complex remote sensing images, current semantic segmentation methods still face challenges and do not perform as well as desired. How to obtain both spatial detail information and semantic information at the same time is an urgent problem to be solved. This letter proposes a context fusion network based on boundary guidance (BGFNet), which incorporates the patch attention module (PAM), the feature maps are enriched with contextual information, improving their ability to capture spatial dependencies. In order to alleviate boundary ambiguity, a boundary guidance module (BGM) is used to weight features with rich semantic boundary information. Furthermore, the compatible fusion module (CFM) is employed to merge high-order and low-order features, creating novel features. Channel attention is then applied to the obtained features allows us to select the desired features by filtering out irrelevant information. We validate our model on the Vaihingen and Potsdam datasets reached 81.65% and 86.94% mean intersection over union (mIoU), respectively, indicating the superiority of the proposed model.
21 January 2025 in The Journal of Supercomputing
11 July 2023 in Research Square
Striving to enhance predictive performance by leveraging auxiliary behaviors, multi-behavior recommendation models have emerged in the realm of e-commerce. These models aim to address the diversity and effectiveness of interactive behaviors. While some methods have shown promising effects, they still exhibit certain limitations, such as overlooking dynamic nature of user interactions. In this paper, we present TKMBR, a multi-behavior recommendation framework based on a temporal knowledge graph in e-commerce. TKMBR incorporates a temporal knowledge graph to capture the temporal dynamics of user behaviors, which allows for the identification of underlying temporal patterns and the capturing of evolving user preferences over time. To augment the understanding of user preferences, heterogeneous signals are integrated and an item-side information knowledge graph is constructed based on various user-item interactions. Moreover, contrastive learning tasks are employed to alleviate the issue of data sparsity. Finally, we evaluate the performance of our approach on two representative recommendation datasets using standard metrics with HR and NDCG. Experimental results demonstrate the effectiveness of TKMBR in improving recommendation quality.
2 March 2023 in Research Square
COVID-19 is spreading globally, posing a great risk and challenging to the world. And wearing a mask can reduce the spread of the virus and the speed of virus transmission, which is an effective mean to combat the spread of coronavirus. To limit the spread of the virus, the requirement to wear a mask is now becoming more common in public places around the world. Nowadays mask wearing detection has become an important task for computer vision to help the global community. The mask detection task involves complex and diverse scenarios with problems of missed and false detection caused by face features being obscured, varying target scales, poor detection of small targets, and small differences in features between correctly and incorrectly worn masks. For this reason, we propose a new mask detection model, MFMDet, which uses Recursive Feature Pyramid to process the multi-scale features extracted by the backbone network, increasing the global features and the perceptual field, and enhancing the detector's ability to adjust to different scales. To ensure the accurate extraction of valid information, we introduce modulated deformable RoIpooling into the detection head to make the network better adapt to the deformation of the target and enhance the spatial and task awareness of the detection head. In addition, we use Joint Image Hybrid Augmentation to increase the number of training samples and diversity to enhance the model generalization ability. Experimental results show that our method improved by 2.5 AP over the baseline, comparing to the baseline on the PWMFD dataset and outperforms other existing target detection algorithms. We also conducted experiments on the WMD dataset to further validate the generalization ability and effectiveness of the proposed method.